Early identification of those at clinical high risk of psychosis (CHR) is critical for maximizing outcomes for those who convert. However, prediction relies largely on subjective symptom reports. Objective biomarkers are essential. My career goal is to use objective computational neuroscience to predict conversion in CHR. In work recently completed with my primary mentor (Dr. Corlett) and published in Science, I examined whether hallucinations might arise from an over-weighting of prior knowledge in perception. We used sensory conditioning to elicit hallucinatory experiences. Participants were exposed to repeated pairings of visual and auditory stimuli and subsequently perceived the auditory stimulus when only the visual was present. We applied this Conditioned Hallucinations paradigm to four groups: participants with psychosis both with (P+H+) and without (P+H-) hallucinations, healthy voice-hearers (P-H+), and healthy controls (P-H-). Conditioned hallucinations were markedly more frequent in those who hallucinate (P+H+ and P-H+) compared with those who do not (P+H-, P-H-). These behavioral data were used to estimate parameters of a Hierarchical Gaussian Filter (HGF) model with the laboratory of Dr. Stephan (co-mentor). Two different model parameters discriminated between groups of individuals with and without auditory hallucinations and, orthogonally, with and without a diagnosable psychotic disorder. On functional imaging analysis, activity in brain regions encoding low-level perceptual belief (e.g., insula, superior temporal sulcus) differentiated those with and without hallucinations. Activity in brain regions encoding change sensitivity (e.g., cerebellum) differentiated those with and without psychosis. These computational and imaging metrics may hasten the detection of conversion in CHR. However, more work is required. We propose 1) to determine whether these markers relate to risk of conversion in CHR; and 2) to determine whether they change with symptom severity over time. This research will provide training in the clinical application of computational models of perception, the evaluation of CHR, and longitudinal data analysis. Our work will be supported by formal didactics and symposia focused on the theory and practice of computational modeling. To meet my career goal, I must understand more deeply how to construct, alter, and utilize computational models of perception so that I may capture the subtle abnormalities of information processing that predate the development of frank hallucinations and psychosis. This proposal will provide me with the additional training and mentored research experiences necessary to become a fully independent investigator who brings the tools of computational neuroscience to the service of the early detection of psychosis.